E2 run e2_laya_rlce_seed44 (epoch_4.pt)
Browse files
e2_laya_rlce_seed44/config.json
ADDED
|
@@ -0,0 +1,75 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": {
|
| 3 |
+
"encoder_name": "answerdotai/ModernBERT-large",
|
| 4 |
+
"head_layers": 2,
|
| 5 |
+
"head_dropout": 0.1,
|
| 6 |
+
"num_qtypes": 3,
|
| 7 |
+
"max_len": 512,
|
| 8 |
+
"head_max_len": 192,
|
| 9 |
+
"k_buckets": [
|
| 10 |
+
2,
|
| 11 |
+
5,
|
| 12 |
+
10
|
| 13 |
+
]
|
| 14 |
+
},
|
| 15 |
+
"training": {
|
| 16 |
+
"dataset_name": "LocalLLaMA/typed-decisions",
|
| 17 |
+
"dataset_config": "all",
|
| 18 |
+
"calib_fraction": 0.1,
|
| 19 |
+
"seed": 44,
|
| 20 |
+
"max_examples": null,
|
| 21 |
+
"ckpt_dir": "checkpoints",
|
| 22 |
+
"result_dir": "results",
|
| 23 |
+
"run_name": "e2_laya_rlce_seed44",
|
| 24 |
+
"epochs": 4,
|
| 25 |
+
"batch_size": 16,
|
| 26 |
+
"accum_steps": 4,
|
| 27 |
+
"num_workers": 4,
|
| 28 |
+
"pin_memory": true,
|
| 29 |
+
"amp": true,
|
| 30 |
+
"amp_dtype": "bf16",
|
| 31 |
+
"gradient_checkpointing": false,
|
| 32 |
+
"lr": 2.5e-05,
|
| 33 |
+
"lr_head": 0.0001,
|
| 34 |
+
"weight_decay": 0.01,
|
| 35 |
+
"grad_clip": 1.0,
|
| 36 |
+
"log_every": 10,
|
| 37 |
+
"eval_every": 1,
|
| 38 |
+
"ckpt_every": 4,
|
| 39 |
+
"w_rl": 1.0,
|
| 40 |
+
"w_ce": 1.0,
|
| 41 |
+
"num_noise_samples": 4,
|
| 42 |
+
"sigma_start": 0.4,
|
| 43 |
+
"sigma_end": 0.1,
|
| 44 |
+
"anneal_sigma": true,
|
| 45 |
+
"reward_w_spherical": 0.75,
|
| 46 |
+
"reward_w_rps": 1.0,
|
| 47 |
+
"arch": {
|
| 48 |
+
"encoder_name": "answerdotai/ModernBERT-large",
|
| 49 |
+
"head_layers": 2,
|
| 50 |
+
"max_len": 512,
|
| 51 |
+
"head_max_len": 192
|
| 52 |
+
},
|
| 53 |
+
"init_from": "convaiinnovations/laya",
|
| 54 |
+
"log_grad_diagnostics": true,
|
| 55 |
+
"lr_scheduler": {
|
| 56 |
+
"type": "cosine",
|
| 57 |
+
"t_max": "auto",
|
| 58 |
+
"eta_min": 1e-06
|
| 59 |
+
},
|
| 60 |
+
"early_stopping": {
|
| 61 |
+
"enabled": false
|
| 62 |
+
},
|
| 63 |
+
"save_best": false,
|
| 64 |
+
"best_metric": "raw_ece",
|
| 65 |
+
"best_mode": "min",
|
| 66 |
+
"wandb": {
|
| 67 |
+
"enabled": true,
|
| 68 |
+
"project": "rlcd-reverse-engineering",
|
| 69 |
+
"entity": null,
|
| 70 |
+
"group": "e2-minimal",
|
| 71 |
+
"log_artifacts": false
|
| 72 |
+
},
|
| 73 |
+
"resume_from": null
|
| 74 |
+
}
|
| 75 |
+
}
|
e2_laya_rlce_seed44/grad_diagnostics.json
ADDED
|
@@ -0,0 +1,308 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"step": 10,
|
| 4 |
+
"epoch": 1,
|
| 5 |
+
"sigma": 0.4,
|
| 6 |
+
"grad_cos_rl_ce": 0.526711106300354,
|
| 7 |
+
"grad_norm_rl": 0.6902408003807068,
|
| 8 |
+
"grad_norm_ce": 0.08752267807722092,
|
| 9 |
+
"grad_norm_ratio_rl_ce": 7.886422302705477
|
| 10 |
+
},
|
| 11 |
+
{
|
| 12 |
+
"step": 20,
|
| 13 |
+
"epoch": 1,
|
| 14 |
+
"sigma": 0.4,
|
| 15 |
+
"grad_cos_rl_ce": 0.7224842309951782,
|
| 16 |
+
"grad_norm_rl": 0.7339413166046143,
|
| 17 |
+
"grad_norm_ce": 0.12021761387586594,
|
| 18 |
+
"grad_norm_ratio_rl_ce": 6.105106339595676
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"step": 30,
|
| 22 |
+
"epoch": 1,
|
| 23 |
+
"sigma": 0.4,
|
| 24 |
+
"grad_cos_rl_ce": 0.7906366586685181,
|
| 25 |
+
"grad_norm_rl": 0.6963644027709961,
|
| 26 |
+
"grad_norm_ce": 0.0969790518283844,
|
| 27 |
+
"grad_norm_ratio_rl_ce": 7.180565180233904
|
| 28 |
+
},
|
| 29 |
+
{
|
| 30 |
+
"step": 40,
|
| 31 |
+
"epoch": 1,
|
| 32 |
+
"sigma": 0.4,
|
| 33 |
+
"grad_cos_rl_ce": 0.6721857786178589,
|
| 34 |
+
"grad_norm_rl": 0.8163612484931946,
|
| 35 |
+
"grad_norm_ce": 0.07112111896276474,
|
| 36 |
+
"grad_norm_ratio_rl_ce": 11.478464630465083
|
| 37 |
+
},
|
| 38 |
+
{
|
| 39 |
+
"step": 50,
|
| 40 |
+
"epoch": 1,
|
| 41 |
+
"sigma": 0.4,
|
| 42 |
+
"grad_cos_rl_ce": 0.7966291904449463,
|
| 43 |
+
"grad_norm_rl": 0.5551742911338806,
|
| 44 |
+
"grad_norm_ce": 0.0913395956158638,
|
| 45 |
+
"grad_norm_ratio_rl_ce": 6.078133884769009
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"step": 60,
|
| 49 |
+
"epoch": 1,
|
| 50 |
+
"sigma": 0.4,
|
| 51 |
+
"grad_cos_rl_ce": 0.735835075378418,
|
| 52 |
+
"grad_norm_rl": 0.6924993991851807,
|
| 53 |
+
"grad_norm_ce": 0.07618748396635056,
|
| 54 |
+
"grad_norm_ratio_rl_ce": 9.089411582236188
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"step": 70,
|
| 58 |
+
"epoch": 1,
|
| 59 |
+
"sigma": 0.4,
|
| 60 |
+
"grad_cos_rl_ce": 0.6695548295974731,
|
| 61 |
+
"grad_norm_rl": 0.5959143042564392,
|
| 62 |
+
"grad_norm_ce": 0.10669995099306107,
|
| 63 |
+
"grad_norm_ratio_rl_ce": 5.584953870271157
|
| 64 |
+
},
|
| 65 |
+
{
|
| 66 |
+
"step": 80,
|
| 67 |
+
"epoch": 1,
|
| 68 |
+
"sigma": 0.4,
|
| 69 |
+
"grad_cos_rl_ce": 0.6584649085998535,
|
| 70 |
+
"grad_norm_rl": 0.604222297668457,
|
| 71 |
+
"grad_norm_ce": 0.1055402085185051,
|
| 72 |
+
"grad_norm_ratio_rl_ce": 5.72504362223725
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"step": 90,
|
| 76 |
+
"epoch": 2,
|
| 77 |
+
"sigma": 0.30000000000000004,
|
| 78 |
+
"grad_cos_rl_ce": 0.6453590989112854,
|
| 79 |
+
"grad_norm_rl": 1.0254168510437012,
|
| 80 |
+
"grad_norm_ce": 0.06525019556283951,
|
| 81 |
+
"grad_norm_ratio_rl_ce": 15.715153681894618
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"step": 100,
|
| 85 |
+
"epoch": 2,
|
| 86 |
+
"sigma": 0.30000000000000004,
|
| 87 |
+
"grad_cos_rl_ce": 0.5302232503890991,
|
| 88 |
+
"grad_norm_rl": 1.332290530204773,
|
| 89 |
+
"grad_norm_ce": 0.06934770941734314,
|
| 90 |
+
"grad_norm_ratio_rl_ce": 19.211745296256044
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"step": 110,
|
| 94 |
+
"epoch": 2,
|
| 95 |
+
"sigma": 0.30000000000000004,
|
| 96 |
+
"grad_cos_rl_ce": 0.6062527894973755,
|
| 97 |
+
"grad_norm_rl": 0.9019922614097595,
|
| 98 |
+
"grad_norm_ce": 0.054473333060741425,
|
| 99 |
+
"grad_norm_ratio_rl_ce": 16.558418784545047
|
| 100 |
+
},
|
| 101 |
+
{
|
| 102 |
+
"step": 120,
|
| 103 |
+
"epoch": 2,
|
| 104 |
+
"sigma": 0.30000000000000004,
|
| 105 |
+
"grad_cos_rl_ce": 0.6394805908203125,
|
| 106 |
+
"grad_norm_rl": 0.9917426109313965,
|
| 107 |
+
"grad_norm_ce": 0.07401520758867264,
|
| 108 |
+
"grad_norm_ratio_rl_ce": 13.39917353799564
|
| 109 |
+
},
|
| 110 |
+
{
|
| 111 |
+
"step": 130,
|
| 112 |
+
"epoch": 2,
|
| 113 |
+
"sigma": 0.30000000000000004,
|
| 114 |
+
"grad_cos_rl_ce": 0.6564541459083557,
|
| 115 |
+
"grad_norm_rl": 0.8749192357063293,
|
| 116 |
+
"grad_norm_ce": 0.058601588010787964,
|
| 117 |
+
"grad_norm_ratio_rl_ce": 14.929957794748933
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"step": 140,
|
| 121 |
+
"epoch": 2,
|
| 122 |
+
"sigma": 0.30000000000000004,
|
| 123 |
+
"grad_cos_rl_ce": 0.6919472217559814,
|
| 124 |
+
"grad_norm_rl": 0.8487017154693604,
|
| 125 |
+
"grad_norm_ce": 0.064200758934021,
|
| 126 |
+
"grad_norm_ratio_rl_ce": 13.219496615944518
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"step": 150,
|
| 130 |
+
"epoch": 2,
|
| 131 |
+
"sigma": 0.30000000000000004,
|
| 132 |
+
"grad_cos_rl_ce": 0.6505917906761169,
|
| 133 |
+
"grad_norm_rl": 1.1652113199234009,
|
| 134 |
+
"grad_norm_ce": 0.06277589499950409,
|
| 135 |
+
"grad_norm_ratio_rl_ce": 18.561444961200564
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"step": 160,
|
| 139 |
+
"epoch": 2,
|
| 140 |
+
"sigma": 0.30000000000000004,
|
| 141 |
+
"grad_cos_rl_ce": 0.44653236865997314,
|
| 142 |
+
"grad_norm_rl": 1.1307870149612427,
|
| 143 |
+
"grad_norm_ce": 0.04861542582511902,
|
| 144 |
+
"grad_norm_ratio_rl_ce": 23.25983976832675
|
| 145 |
+
},
|
| 146 |
+
{
|
| 147 |
+
"step": 170,
|
| 148 |
+
"epoch": 2,
|
| 149 |
+
"sigma": 0.30000000000000004,
|
| 150 |
+
"grad_cos_rl_ce": 0.664006233215332,
|
| 151 |
+
"grad_norm_rl": 0.9291825294494629,
|
| 152 |
+
"grad_norm_ce": 0.03998361527919769,
|
| 153 |
+
"grad_norm_ratio_rl_ce": 23.239082383150315
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"step": 180,
|
| 157 |
+
"epoch": 3,
|
| 158 |
+
"sigma": 0.2,
|
| 159 |
+
"grad_cos_rl_ce": 0.6483744382858276,
|
| 160 |
+
"grad_norm_rl": 1.6389068365097046,
|
| 161 |
+
"grad_norm_ce": 0.049361273646354675,
|
| 162 |
+
"grad_norm_ratio_rl_ce": 33.202280156941164
|
| 163 |
+
},
|
| 164 |
+
{
|
| 165 |
+
"step": 190,
|
| 166 |
+
"epoch": 3,
|
| 167 |
+
"sigma": 0.2,
|
| 168 |
+
"grad_cos_rl_ce": 0.8396736979484558,
|
| 169 |
+
"grad_norm_rl": 1.4482805728912354,
|
| 170 |
+
"grad_norm_ce": 0.04289504885673523,
|
| 171 |
+
"grad_norm_ratio_rl_ce": 33.763350584547275
|
| 172 |
+
},
|
| 173 |
+
{
|
| 174 |
+
"step": 200,
|
| 175 |
+
"epoch": 3,
|
| 176 |
+
"sigma": 0.2,
|
| 177 |
+
"grad_cos_rl_ce": 0.8082038164138794,
|
| 178 |
+
"grad_norm_rl": 1.1133170127868652,
|
| 179 |
+
"grad_norm_ce": 0.05074053630232811,
|
| 180 |
+
"grad_norm_ratio_rl_ce": 21.941372597115876
|
| 181 |
+
},
|
| 182 |
+
{
|
| 183 |
+
"step": 210,
|
| 184 |
+
"epoch": 3,
|
| 185 |
+
"sigma": 0.2,
|
| 186 |
+
"grad_cos_rl_ce": 0.6891321539878845,
|
| 187 |
+
"grad_norm_rl": 1.4620789289474487,
|
| 188 |
+
"grad_norm_ce": 0.03497035428881645,
|
| 189 |
+
"grad_norm_ratio_rl_ce": 41.80909683877646
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"step": 220,
|
| 193 |
+
"epoch": 3,
|
| 194 |
+
"sigma": 0.2,
|
| 195 |
+
"grad_cos_rl_ce": 0.8432408571243286,
|
| 196 |
+
"grad_norm_rl": 1.5246686935424805,
|
| 197 |
+
"grad_norm_ce": 0.03936393931508064,
|
| 198 |
+
"grad_norm_ratio_rl_ce": 38.732624835603474
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"step": 230,
|
| 202 |
+
"epoch": 3,
|
| 203 |
+
"sigma": 0.2,
|
| 204 |
+
"grad_cos_rl_ce": 0.7183141708374023,
|
| 205 |
+
"grad_norm_rl": 1.5086736679077148,
|
| 206 |
+
"grad_norm_ce": 0.043949566781520844,
|
| 207 |
+
"grad_norm_ratio_rl_ce": 34.32738428134076
|
| 208 |
+
},
|
| 209 |
+
{
|
| 210 |
+
"step": 240,
|
| 211 |
+
"epoch": 3,
|
| 212 |
+
"sigma": 0.2,
|
| 213 |
+
"grad_cos_rl_ce": 0.7045532464981079,
|
| 214 |
+
"grad_norm_rl": 1.3480327129364014,
|
| 215 |
+
"grad_norm_ce": 0.04354315251111984,
|
| 216 |
+
"grad_norm_ratio_rl_ce": 30.958546526739127
|
| 217 |
+
},
|
| 218 |
+
{
|
| 219 |
+
"step": 250,
|
| 220 |
+
"epoch": 3,
|
| 221 |
+
"sigma": 0.2,
|
| 222 |
+
"grad_cos_rl_ce": 0.6744354963302612,
|
| 223 |
+
"grad_norm_rl": 1.3695229291915894,
|
| 224 |
+
"grad_norm_ce": 0.04818682745099068,
|
| 225 |
+
"grad_norm_ratio_rl_ce": 28.421105966863838
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"step": 260,
|
| 229 |
+
"epoch": 4,
|
| 230 |
+
"sigma": 0.09999999999999998,
|
| 231 |
+
"grad_cos_rl_ce": 0.5611343383789062,
|
| 232 |
+
"grad_norm_rl": 3.6783852577209473,
|
| 233 |
+
"grad_norm_ce": 0.04595884680747986,
|
| 234 |
+
"grad_norm_ratio_rl_ce": 80.03650033103715
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"step": 270,
|
| 238 |
+
"epoch": 4,
|
| 239 |
+
"sigma": 0.09999999999999998,
|
| 240 |
+
"grad_cos_rl_ce": 0.7897813320159912,
|
| 241 |
+
"grad_norm_rl": 3.0842111110687256,
|
| 242 |
+
"grad_norm_ce": 0.05890565365552902,
|
| 243 |
+
"grad_norm_ratio_rl_ce": 52.35849056365126
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"step": 280,
|
| 247 |
+
"epoch": 4,
|
| 248 |
+
"sigma": 0.09999999999999998,
|
| 249 |
+
"grad_cos_rl_ce": 0.6877776384353638,
|
| 250 |
+
"grad_norm_rl": 3.217217445373535,
|
| 251 |
+
"grad_norm_ce": 0.03197171166539192,
|
| 252 |
+
"grad_norm_ratio_rl_ce": 100.62700048856134
|
| 253 |
+
},
|
| 254 |
+
{
|
| 255 |
+
"step": 290,
|
| 256 |
+
"epoch": 4,
|
| 257 |
+
"sigma": 0.09999999999999998,
|
| 258 |
+
"grad_cos_rl_ce": 0.712121844291687,
|
| 259 |
+
"grad_norm_rl": 3.197781562805176,
|
| 260 |
+
"grad_norm_ce": 0.03381294757127762,
|
| 261 |
+
"grad_norm_ratio_rl_ce": 94.57269455921461
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"step": 300,
|
| 265 |
+
"epoch": 4,
|
| 266 |
+
"sigma": 0.09999999999999998,
|
| 267 |
+
"grad_cos_rl_ce": 0.6289786696434021,
|
| 268 |
+
"grad_norm_rl": 3.1438448429107666,
|
| 269 |
+
"grad_norm_ce": 0.04466373473405838,
|
| 270 |
+
"grad_norm_ratio_rl_ce": 70.38920640269306
|
| 271 |
+
},
|
| 272 |
+
{
|
| 273 |
+
"step": 310,
|
| 274 |
+
"epoch": 4,
|
| 275 |
+
"sigma": 0.09999999999999998,
|
| 276 |
+
"grad_cos_rl_ce": 0.714044988155365,
|
| 277 |
+
"grad_norm_rl": 3.150677442550659,
|
| 278 |
+
"grad_norm_ce": 0.04863156005740166,
|
| 279 |
+
"grad_norm_ratio_rl_ce": 64.78668253356044
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"step": 320,
|
| 283 |
+
"epoch": 4,
|
| 284 |
+
"sigma": 0.09999999999999998,
|
| 285 |
+
"grad_cos_rl_ce": 0.6874269247055054,
|
| 286 |
+
"grad_norm_rl": 3.037999153137207,
|
| 287 |
+
"grad_norm_ce": 0.05100342258810997,
|
| 288 |
+
"grad_norm_ratio_rl_ce": 59.56461349018237
|
| 289 |
+
},
|
| 290 |
+
{
|
| 291 |
+
"step": 330,
|
| 292 |
+
"epoch": 4,
|
| 293 |
+
"sigma": 0.09999999999999998,
|
| 294 |
+
"grad_cos_rl_ce": 0.7293472290039062,
|
| 295 |
+
"grad_norm_rl": 3.3383219242095947,
|
| 296 |
+
"grad_norm_ce": 0.02300458401441574,
|
| 297 |
+
"grad_norm_ratio_rl_ce": 145.11550924449003
|
| 298 |
+
},
|
| 299 |
+
{
|
| 300 |
+
"step": 340,
|
| 301 |
+
"epoch": 4,
|
| 302 |
+
"sigma": 0.09999999999999998,
|
| 303 |
+
"grad_cos_rl_ce": 0.7144708037376404,
|
| 304 |
+
"grad_norm_rl": 3.679173469543457,
|
| 305 |
+
"grad_norm_ce": 0.046368807554244995,
|
| 306 |
+
"grad_norm_ratio_rl_ce": 79.3458720118981
|
| 307 |
+
}
|
| 308 |
+
]
|
e2_laya_rlce_seed44/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:78fe096caa2d1b1c8c907056340bfe48c0d6fbce0861fd409f162fc6047de903
|
| 3 |
+
size 1684141044
|
e2_laya_rlce_seed44/model_config.json
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"encoder_name": "answerdotai/ModernBERT-large",
|
| 3 |
+
"head_layers": 2,
|
| 4 |
+
"head_dropout": 0.1,
|
| 5 |
+
"num_qtypes": 3,
|
| 6 |
+
"max_len": 512,
|
| 7 |
+
"head_max_len": 192,
|
| 8 |
+
"k_buckets": [
|
| 9 |
+
2,
|
| 10 |
+
5,
|
| 11 |
+
10
|
| 12 |
+
],
|
| 13 |
+
"source_checkpoint": "epoch_4.pt",
|
| 14 |
+
"epoch": 4,
|
| 15 |
+
"step": 340
|
| 16 |
+
}
|
e2_laya_rlce_seed44/test_eval.json
ADDED
|
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"raw": {
|
| 3 |
+
"ece": 0.13394594296813012,
|
| 4 |
+
"brier": 0.05498596685757368,
|
| 5 |
+
"nll": 0.8672405917616561,
|
| 6 |
+
"accuracy": 0.7715,
|
| 7 |
+
"soft_accuracy": 0.516161304473877,
|
| 8 |
+
"mean_confidence": 0.6375540485084057,
|
| 9 |
+
"score_mae": 0.22897809116868303,
|
| 10 |
+
"n": 2000,
|
| 11 |
+
"by_type": {
|
| 12 |
+
"choice": {
|
| 13 |
+
"ece": 0.12707374493281046,
|
| 14 |
+
"brier": 0.06884223616547691,
|
| 15 |
+
"nll": 0.9961476571982105,
|
| 16 |
+
"accuracy": 0.7433333333333333,
|
| 17 |
+
"soft_accuracy": 0.4729881009707848,
|
| 18 |
+
"mean_confidence": 0.6162595846255621,
|
| 19 |
+
"score_mae": null,
|
| 20 |
+
"n": 600
|
| 21 |
+
},
|
| 22 |
+
"score": {
|
| 23 |
+
"ece": 0.17143568731844427,
|
| 24 |
+
"brier": 0.057610363487638094,
|
| 25 |
+
"nll": 1.0435749628953637,
|
| 26 |
+
"accuracy": 0.73875,
|
| 27 |
+
"soft_accuracy": 0.4323799050040543,
|
| 28 |
+
"mean_confidence": 0.5673143507540226,
|
| 29 |
+
"score_mae": 0.22897809116868303,
|
| 30 |
+
"n": 800
|
| 31 |
+
},
|
| 32 |
+
"noul": {
|
| 33 |
+
"ece": 0.09800974751512209,
|
| 34 |
+
"brier": 0.037630502042917915,
|
| 35 |
+
"nll": 0.5032210314801584,
|
| 36 |
+
"accuracy": 0.8433333333333334,
|
| 37 |
+
"soft_accuracy": 0.6710430406033993,
|
| 38 |
+
"mean_confidence": 0.7525014427304267,
|
| 39 |
+
"score_mae": null,
|
| 40 |
+
"n": 600
|
| 41 |
+
}
|
| 42 |
+
}
|
| 43 |
+
},
|
| 44 |
+
"post_temperature": {
|
| 45 |
+
"ece": 0.1504889203310013,
|
| 46 |
+
"brier": 0.0535475225614009,
|
| 47 |
+
"nll": 0.864496507877484,
|
| 48 |
+
"accuracy": 0.7715,
|
| 49 |
+
"soft_accuracy": 0.5082126140743494,
|
| 50 |
+
"mean_confidence": 0.6210110761225224,
|
| 51 |
+
"score_mae": 0.23021191493957305,
|
| 52 |
+
"n": 2000,
|
| 53 |
+
"by_type": {
|
| 54 |
+
"choice": {
|
| 55 |
+
"ece": 0.15238359754284223,
|
| 56 |
+
"brier": 0.06618621635801294,
|
| 57 |
+
"nll": 0.9919972387328744,
|
| 58 |
+
"accuracy": 0.7433333333333333,
|
| 59 |
+
"soft_accuracy": 0.460888055190444,
|
| 60 |
+
"mean_confidence": 0.5909497653941314,
|
| 61 |
+
"score_mae": null,
|
| 62 |
+
"n": 600
|
| 63 |
+
},
|
| 64 |
+
"score": {
|
| 65 |
+
"ece": 0.18502540819346902,
|
| 66 |
+
"brier": 0.05639494055324576,
|
| 67 |
+
"nll": 1.040980128934607,
|
| 68 |
+
"accuracy": 0.73875,
|
| 69 |
+
"soft_accuracy": 0.4262336232326925,
|
| 70 |
+
"mean_confidence": 0.5537245621159673,
|
| 71 |
+
"score_mae": 0.23021191493957305,
|
| 72 |
+
"n": 800
|
| 73 |
+
},
|
| 74 |
+
"noul": {
|
| 75 |
+
"ece": 0.10953281487027804,
|
| 76 |
+
"brier": 0.037112271442329016,
|
| 77 |
+
"nll": 0.501684282279263,
|
| 78 |
+
"accuracy": 0.8433333333333334,
|
| 79 |
+
"soft_accuracy": 0.664842494080464,
|
| 80 |
+
"mean_confidence": 0.7407877388596534,
|
| 81 |
+
"score_mae": null,
|
| 82 |
+
"n": 600
|
| 83 |
+
}
|
| 84 |
+
}
|
| 85 |
+
},
|
| 86 |
+
"fitted_temperature": {
|
| 87 |
+
"type1_bucket1": 1.0628252029418945,
|
| 88 |
+
"type0_bucket1": 1.108896255493164,
|
| 89 |
+
"type2_bucket0": 1.0960332155227661
|
| 90 |
+
}
|
| 91 |
+
}
|
e2_laya_rlce_seed44/train_log.json
ADDED
|
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"epoch": 1,
|
| 4 |
+
"step": 85,
|
| 5 |
+
"loss": 0.7647291334236369,
|
| 6 |
+
"sigma": 0.4,
|
| 7 |
+
"epoch_seconds": 78.93924188613892,
|
| 8 |
+
"train_rows": 5400,
|
| 9 |
+
"raw_ece": 0.1340081213414669,
|
| 10 |
+
"raw_brier": 0.10269942383555446,
|
| 11 |
+
"raw_nll": 0.930330185526982,
|
| 12 |
+
"raw_accuracy": 0.7216666666666667,
|
| 13 |
+
"raw_soft_accuracy": 0.47850164763008557,
|
| 14 |
+
"raw_mean_confidence": 0.5899476708471775,
|
| 15 |
+
"raw_score_mae": 0.3357344947056845,
|
| 16 |
+
"raw_n": 600
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 2,
|
| 20 |
+
"step": 170,
|
| 21 |
+
"loss": 0.6586984009427183,
|
| 22 |
+
"sigma": 0.30000000000000004,
|
| 23 |
+
"epoch_seconds": 80.43961930274963,
|
| 24 |
+
"train_rows": 5400,
|
| 25 |
+
"raw_ece": 0.10455580711364744,
|
| 26 |
+
"raw_brier": 0.07312167295666323,
|
| 27 |
+
"raw_nll": 0.8740207870521893,
|
| 28 |
+
"raw_accuracy": 0.765,
|
| 29 |
+
"raw_soft_accuracy": 0.5270250970870256,
|
| 30 |
+
"raw_mean_confidence": 0.6620155566930771,
|
| 31 |
+
"raw_score_mae": 0.2484339059330523,
|
| 32 |
+
"raw_n": 600
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 3,
|
| 36 |
+
"step": 255,
|
| 37 |
+
"loss": 0.6373752018108088,
|
| 38 |
+
"sigma": 0.2,
|
| 39 |
+
"epoch_seconds": 81.76218628883362,
|
| 40 |
+
"train_rows": 5400,
|
| 41 |
+
"raw_ece": 0.13365768944223722,
|
| 42 |
+
"raw_brier": 0.06672702780118289,
|
| 43 |
+
"raw_nll": 0.8589709452011933,
|
| 44 |
+
"raw_accuracy": 0.7583333333333333,
|
| 45 |
+
"raw_soft_accuracy": 0.5127109212179979,
|
| 46 |
+
"raw_mean_confidence": 0.6268951519330342,
|
| 47 |
+
"raw_score_mae": 0.24661954320035875,
|
| 48 |
+
"raw_n": 600
|
| 49 |
+
},
|
| 50 |
+
{
|
| 51 |
+
"epoch": 4,
|
| 52 |
+
"step": 340,
|
| 53 |
+
"loss": 0.7021594749654041,
|
| 54 |
+
"sigma": 0.09999999999999998,
|
| 55 |
+
"epoch_seconds": 81.0947036743164,
|
| 56 |
+
"train_rows": 5400,
|
| 57 |
+
"raw_ece": 0.13961881279945373,
|
| 58 |
+
"raw_brier": 0.062124831366484064,
|
| 59 |
+
"raw_nll": 0.851531696934253,
|
| 60 |
+
"raw_accuracy": 0.7683333333333333,
|
| 61 |
+
"raw_soft_accuracy": 0.5189536493892472,
|
| 62 |
+
"raw_mean_confidence": 0.6349892205993335,
|
| 63 |
+
"raw_score_mae": 0.23095368212088943,
|
| 64 |
+
"raw_n": 600
|
| 65 |
+
}
|
| 66 |
+
]
|